What Resistbotters Are Writing to Trump

Its been two weeks since Resistbot enabled Americans to send messages directly to the White House and with news of indictments related to the Russia scandal dropping this week, we thought it would be a good time to see what Resistbotters have written to Trump.

We analyzed roughly 2,500 messages directed to the White House (anonymized to protect user privacy) and looked at the most frequently used words and topics referenced in the past two weeks. We also analyzed the sentiment of these messages to get a sense of how much heat the president has received from Resistbot.

First, the most frequently used words:

The fourth most frequently used word is resign, appearing 635 times and in nearly 20% of all messages sent to the president to date. We fitted all 2,500 messages to a topic model via latent Dirichlet allocation (LDA). The model resolved 5 major themes in the messages:

1. resign e.g. “You need to resign immediately.” Messages in this group often reference Trump’s disgraceful feud with war widow Myeshia Johnson and Rep. Frederica Wilson.
2. puerto rico e.g. “Puerto Ricans are American citizens!” In the aftermath of the hurricanes, Trump’s callous response to the humanitarian crisis in Puerto Rico has drawn the ire of Resistbotters.
3. healthcare e.g. “Don’t take away our healthcare!” Roughly half of all messages sent to elected officials via Resistbot mention healthcare, and Trump is not immune to this pressure. Additionally, messages in this topic also mention Medicaid, mental health, and gun control.
4. russia e.g. “Enforce the Russia Sanctions!” Messages in this topic cover many foreign policy issues, like Iran and North Korea, but Trump delaying enforcement of sanctions on Russia is the most salient.
5. net neutrality e.g. “Please support net neutrality.”

Now that we have a sense of what Resistbotters are writing to Trump, we can take a look at how they sound. To do this, we analyzed the sentiment of every message using the Finn Årup Nielsen (AFINN) sentiment model. This model assigns English words a sentiment score, ranging from -5 to +5. Words with negative values have a negative sentiment, while words with positive values have a positive sentiment. For example, a word like bastard has a score of -5, while a word like outstanding has a score of +5. Most English words tend to be pretty neutral in this model, like demand (-1) or protect (+1).

We grouped every word by the sender’s state, computed the average sentiment score, and produced a ranking of states from most negative to most positive sentiment.

Minnesotans may be super-nice in every other facet of life, but when it comes to writing messages to Trump, they’re the meanest — with an average sentiment score of -0.51. Meanwhile, Virginians have been writing the most polite messages to the president, with an average sentiment score of +0.17.

In general, users with Republican representatives and senators tend to write more negative messages to Trump, with an average sentiment score of -0.24. Users with Democratic representatives and senators tend to write more positive messages (relatively speaking), with an average sentiment score of -0.10.

It’s worth noting that this method of sentiment analysis doesn’t do a great job of picking up on sarcasm. The following message has a sentiment score of +3.

Today is a great day for your resignation!

Not surprisingly, many Resistbot messages to Trump are pretty sarcastic. If you want to make your voice heard, click here or text RESIST to 50409.